(Bloomberg) -- Juergen Schmidhuber taught a computer to park a car. He’s also showing that same machine how to trade stocks and detect flaws in steel production. Unrelated as these tasks may appear, Schmidhuber thinks a seemingly random training regimen is key to creating artificial intelligence that can solve any problem.
Schmidhuber’s AI theories tend to carry weight. In 1997, he co-authored a seminal paper that laid the groundwork for modern AI systems. It introduced a form of memory modeled after the human brain that helps computers make decisions based on context and prior experiences. The concept underpins most modern AI networks. Today he serves as the head of a Swiss AI institute that’s proven to be a fertile breeding ground for many of the world’s top experts. The New York Times recently referred to him as a would-be father of AI.
In the last couple years, Schmidhuber has been toiling away in relative obscurity on a small venture called Nnaisense SA. As his ambitions grew, the company, whose name is pronounced “nascence,” raised its first round of venture capital this month. The funding shows that investors are willing to make a bet on Schmidhuber and his big idea—that just as with a child, the more skills Nnaisense’s AI acquires, the easier it is to learn new ones.
Schmidhuber, a consummate academic, said he decided to raise money partly to guard against the likes of Alphabet Inc., Apple Inc., Amazon.com Inc. and Baidu Inc. “The big companies came in and tried to hire my best people away,” he said. “One of the reasons that I had to become an entrepreneur, as well as an academic, was to try to defend that.”
Creating an all-purpose AI is a steep challenge for any company, let alone one with 15 employees. Nnaisense may not have the vast reams of search data on which Google’s machines learn, but it can more nimbly jump from project to project, exposing its digital mind to many different experiences in the process. “We always let it learn something new on top of the previous professions,” Schmidhuber said. “As you are learning new things, you want to make the task more complex.”
Schmidhuber believes that by contracting out the services of his company to a variety of industries, the skills learned from each job will gradually snowball into a system capable of tackling any challenge it faces. Some within the AI community are skeptical of Schmidhuber’s bullishness. Companies such as International Business Machines Corp. and German startup Arago GmbH have worked for years on applying AI to various unrelated projects without achieving consciousness. (But Watson has gotten pretty adept at chess and Jeopardy!)
Nnaisense’s new financier is Madrid’s Alma Mundi Ventures. Rajeev Singh-Molares, a partner at the firm, said he’s optimistic about the startup’s chances going up against much larger companies. His firm invested $2 million to $5 million, though he declined to specify the exact sum. “The big boys and girls have a lot of data, but they don’t have this cross-industry data—in insurance, banking, automotive, industrial processes and a couple of others,” he said.
Nnaisense recently worked with Volkswagen AG’s Audi luxury auto unit to create a miniaturized car able to park itself. The system used cameras to teach itself how to drive on its own. Other autonomous driving rigs rely on many other sensors, including Lidar and radar, to find their way around based on predetermined parameters. Nnaisense applied similar techniques for its work with global steel giant ArcelorMittal and for the investing program, with hedge fund Acatis Investment GmbH.
Commercial AI development within Alphabet, Apple, Facebook and elsewhere is typically determined by which endeavors are likely to be the most lucrative. Nnaisense said it chooses projects based on whether they’ll benefit the machine’s knowledge, not which will bring in the highest fees. The company generated well under 10 million euros ($11 million) last year but expects to at least double its sales this year.
Faustino Gomez, Nnaisense’s chief executive officer who started the company with Schmidhuber and three others, said they’re encouraged by early signs of the machine’s ability to transfer knowledge learned from one task to the next: “What we ultimately want to do as we move forward is not to have to learn a solution for every problem from scratch.”
Register or login for access to this item and much more
All Information Management content is archived after seven days.
Community members receive:
- All recent and archived articles
- Conference offers and updates
- A full menu of enewsletter options
- Web seminars, white papers, ebooks
Already have an account? Log In
Don't have an account? Register for Free Unlimited Access